<![CDATA[This research aims to perform feature extraction on the dynamic characteristics on dynamic characteristic data on multisensor data. The training data consists of aroma recordings from 6 different species of mint at the Botanical Institute of Karlsruhe Institute of Technology (KIT), Germany, recorded using 12 different sensors. The dataset consists of 28,746 data points collected over a period of 175.52 minutes. The data underwent preprocessing to address spiking issues and standardize the data. Three testing stages were conducted, which are Performance testing of Feature Extraction using raw data, Performance testing of Feature Extraction using Piecewise Linear Regression (PLR) Data, and Performance testing of a Classification Model using the k-Nearest Neighbor (k-NN) algorithm. Feature Extraction was performed using the PCA technique to obtain Principal Component (PC) values of the reduced-dimensional data. For the Feature Extraction using raw data, PC1 had a value of 96.35% and PC2 had a value of 1.84%. Meanwhile, for the Feature Extraction using 12 PLR data, PC1 had a value of 95.95% and PC2 had a value of 1.88%. And for the Feature Extraction using 24 PLR data, PC1 had a value of 95.77% and PC2 had a value of 1.75%. The evaluation testing of the Feature Extraction results employed a machine learning model with the k-NN method. The training results showed that the k-NN model using raw data before PCA Feature Extraction achieved an Accuracy of 95.67% with a computation time of 0.3 seconds, and after PCA Feature Extraction, it achieved an Accuracy of 98.7% with a computation time of 0.19 seconds. In contrast, the k-NN model with 12 PLR Pre-processing before PCA Feature Extraction obtained an Accuracy of 56.67% with a computation time of 0.049 seconds, and after PCA Feature Extraction, it achieved an Accuracy of 85.83% with a computation time of 0.041seconds. Similarly, the k-NN model with 24 PLR Pre-processing before PCA Feature Extraction obtained an Accuracy of 61.67% with a computation time of 0.071 seconds, and after PCA Feature Extraction, it achieved an Accuracy of 90.21% with a computation time of 0.063 seconds. These results indicate that PCA Feature Extraction successfully improved the Accuracy of the prediction model, even when the data dimensions were reduced. The development of this system can serve as an alternative for various data analysis and machine learning algorithms. Keywords: Multisensor, Quartz Cristal Microbalance (QCM), Principal Component Analysis (PCA)]]>
Copyrights © 2023
